Consistent binding forms

[wenkokke]

The grammar for special binding forms currently results in inconsistent parse trees and include the opening parentheses in the name. This leads to several usability issues:

• Alias key and sleep to actions #13
• support no space between spoken form and parens? #21
• More idiomatic parse for face, gamepad, etc #31

This issue tracks those problems, as they share a common cause and solution.

[wenkokke]

The grammar doesn’t support command bindings without a space after the first word, such as:

hello(world):
  …

See #21

[wenkokke]

The parser trees for special binding forms include the opening parenthesis in the name of the binding, as they’re parser as a single token, e.g.,

parrot(click):
  …

…results in a special binding with the name parrot(.

See #31 and #13

[wenkokke]

As mentioned elsewhere, implementing this as an ambiguous grammar with higher priorities for the special binding forms results in strange error correcting behaviour. I believe this is because tree-sitter doesn’t backtrack, and therefore it commits to, e.g., a key binding form whenever it sees something starting with the word “key”, and then error corrects to something strange if we really dealing with a command declaration, e.g., key enter: … gets parsed as a key binding with an erroneous extra word “enter” and a missing argument.

[wenkokke]

The best solution is probably to implement this using lookahead in the parser, i.e., if we see the name of a special binding form at the start of the line, and it has a well formed argument list followed by a colon, we emit the name wrapped in a special_binding token.

Implementing this in the lexer requires modifying scanner.cc to add the special binding logic on top of the matches header logic, and probably requires writing a basic lexing framework—using nested if statements we would end up having to nest as deep as the length of the longest special binding name, which is currently settings… which would be a nightmare to write and to maintain.